Synergistic Entropy Coding and Quantization Enable Efficient on Device Neural Networks

Synergistic Entropy Coding and Quantization Enable Efficient on Device Neural Networks | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Synergistic Entropy Coding and Quantization Enable Efficient on Device Neural Networks Abulfadhel Amer Saihood Altufaili, Dunya Mohammed Shleej This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8018121/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Deploying deep neural networks (DNNs) on edge devices poses significant challenges due to constrained memory, compute, and energy resources. Conventional model compression pipelines—comprising separate stages of pruning, quantization, and entropy coding—often fail to deliver optimal trade-offs between efficiency and accuracy. In this work, we propose Synergistic Entropy–Quantization (SyE-C²Q), a unified compression framework that jointly optimizes quantization precision and entropy coding based on the statistical structure of model parameters. By aligning quantization levels with symbol probabilities and adapting quantization step sizes to entropy estimates, SyE-C²Q reduces the bit-rate while maintaining high inference accuracy. Extensive experiments on MobileNet-V2 and ResNet-18 with CIFAR-10 and ImageNet demonstrate that SyE-C²Q achieves up to 3.6× model compression, < 1% accuracy degradation, and ~ 40% energy savings compared to conventional post-training quantization techniques. Furthermore, the compressed models exhibit improved inference latency and memory utilization on ARM-based edge hardware. Unlike traditional pipelines, SyE-C²Q integrates entropy-guided quantization directly into the compression loop, establishing a new benchmark in the design of resource-efficient, deployable deep learning systems. On-Device AI Edge AI Neural Network Compression Quantization Entropy Coding Rate–Distortion Optimization Energy-Efficient Inference Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8018121","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":554456156,"identity":"c324c98d-d540-4326-87b6-4a5209740f5f","order_by":0,"name":"Abulfadhel Amer Saihood Altufaili","email":"data:image/png;base64,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","orcid":"","institution":"Alfurat Al-Awsat Technical University","correspondingAuthor":true,"prefix":"","firstName":"Abulfadhel","middleName":"Amer Saihood","lastName":"Altufaili","suffix":""},{"id":554456158,"identity":"ac27056f-ab3b-4437-9b1b-1d6dbc802491","order_by":1,"name":"Dunya Mohammed Shleej","email":"","orcid":"","institution":"University of Imam Jaafar Sadiq","correspondingAuthor":false,"prefix":"","firstName":"Dunya","middleName":"Mohammed","lastName":"Shleej","suffix":""}],"badges":[],"createdAt":"2025-11-03 10:38:40","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8018121/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8018121/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":97668197,"identity":"858ffb5a-433c-409b-8eea-bfb720d10708","added_by":"auto","created_at":"2025-12-08 09:25:02","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2025597,"visible":true,"origin":"","legend":"","description":"","filename":"SynergisticEntropyCodingandQuantizationEnableEfficientOnDeviceNeuralNetworks.docx","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/93c9930c85c95382711173fc.docx"},{"id":97475052,"identity":"b08e83f7-39dd-4a3b-b144-0ef4677415f8","added_by":"auto","created_at":"2025-12-04 18:40:02","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5123,"visible":true,"origin":"","legend":"","description":"","filename":"e6d2fd47c8da4a9ea38682ee21f1cb4d.json","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/02e7fc53be3fad52bb736739.json"},{"id":97475063,"identity":"2f083cc8-e0f8-4c0b-a17f-df1bad248c26","added_by":"auto","created_at":"2025-12-04 18:40:03","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":73757,"visible":true,"origin":"","legend":"","description":"","filename":"e6d2fd47c8da4a9ea38682ee21f1cb4d1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/0b257c9c2e8f83adccf90033.xml"},{"id":97475059,"identity":"092f7d63-fe97-4bb7-9547-eaabefc2a6df","added_by":"auto","created_at":"2025-12-04 18:40:03","extension":"eps","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":283433,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/c23a3fca7be78517df773092.eps"},{"id":97475053,"identity":"e0080ec8-2cc4-4c91-ae07-76c6da88f42f","added_by":"auto","created_at":"2025-12-04 18:40:03","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":474538,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/466f468a622936a459265dc3.png"},{"id":97668971,"identity":"51399808-8e72-4200-8dd6-bb07697644d6","added_by":"auto","created_at":"2025-12-08 09:26:53","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":518039,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/cbd6b0ce9cf5b0de5711a3a1.png"},{"id":97475064,"identity":"709bfc88-bbca-44e9-94f3-3eedebb58ebc","added_by":"auto","created_at":"2025-12-04 18:40:03","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":461108,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/740f1f05e3e36045565c0c55.png"},{"id":97475056,"identity":"bf70ff68-c325-43aa-93a3-aff71b1a1470","added_by":"auto","created_at":"2025-12-04 18:40:03","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":527012,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/d5ce8b2c819c291864a51c7a.png"},{"id":97475054,"identity":"9428fbaa-e8d3-42f8-aafc-90cb81753791","added_by":"auto","created_at":"2025-12-04 18:40:03","extension":"jpeg","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1074,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/a11b2c7d2472975a8c668b6f.jpeg"},{"id":97669076,"identity":"1343ae5f-0f38-4741-a495-0c2ffb6e8ddd","added_by":"auto","created_at":"2025-12-08 09:27:10","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33187,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/8f2249d71c4be82f616aab86.png"},{"id":97670384,"identity":"1e4ede28-a178-4d5a-b5ed-d0759ea94a62","added_by":"auto","created_at":"2025-12-08 09:30:28","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":23169,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/6c1542d854a088d7a65662d2.png"},{"id":97669865,"identity":"3d894baf-1669-48be-9e9a-4bce0562789c","added_by":"auto","created_at":"2025-12-08 09:29:11","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":24232,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/c00b784ece6463ada119fffc.png"},{"id":97668832,"identity":"84c537ae-3869-45f6-96a4-bff161416757","added_by":"auto","created_at":"2025-12-08 09:26:21","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":23068,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/622cd4eb5bdf549b7ac00b8d.png"},{"id":97475068,"identity":"25bf3791-d911-4f71-8989-c68acaaa6e24","added_by":"auto","created_at":"2025-12-04 18:40:03","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":935,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/de03b9f1a9f281bfe6d22ef9.png"},{"id":97669377,"identity":"950bf860-9481-4b7f-8946-e790098b615d","added_by":"auto","created_at":"2025-12-08 09:27:54","extension":"xml","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":69611,"visible":true,"origin":"","legend":"","description":"","filename":"e6d2fd47c8da4a9ea38682ee21f1cb4d1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/2d2711718db23a5e45d5094a.xml"},{"id":97475066,"identity":"dc3ac982-63b4-4762-93b6-89b858ffff81","added_by":"auto","created_at":"2025-12-04 18:40:03","extension":"html","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":81588,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1/ad5696cc489d2c47711850ae.html"},{"id":105221389,"identity":"3f68b43b-ab59-4b75-9e34-24577a9f7373","added_by":"auto","created_at":"2026-03-23 15:42:02","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":600936,"visible":true,"origin":"","legend":"","description":"","filename":"SynergisticEntropyCodingandQuantizationEnableEfficientOnDeviceNeuralNetworks.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8018121/v1_covered_5ffe5f40-cf5f-4b40-9df7-32efa799b194.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Synergistic Entropy Coding and Quantization Enable Efficient on Device Neural Networks","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"On-Device AI, Edge AI, Neural Network Compression, Quantization, Entropy Coding, Rate–Distortion Optimization, Energy-Efficient Inference","lastPublishedDoi":"10.21203/rs.3.rs-8018121/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8018121/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDeploying deep neural networks (DNNs) on edge devices poses significant challenges due to constrained memory, compute, and energy resources. Conventional model compression pipelines\u0026mdash;comprising separate stages of pruning, quantization, and entropy coding\u0026mdash;often fail to deliver optimal trade-offs between efficiency and accuracy. In this work, we propose Synergistic Entropy\u0026ndash;Quantization (SyE-C\u0026sup2;Q), a unified compression framework that jointly optimizes quantization precision and entropy coding based on the statistical structure of model parameters. By aligning quantization levels with symbol probabilities and adapting quantization step sizes to entropy estimates, SyE-C\u0026sup2;Q reduces the bit-rate while maintaining high inference accuracy.\u003c/p\u003e\u003cp\u003eExtensive experiments on MobileNet-V2 and ResNet-18 with CIFAR-10 and ImageNet demonstrate that SyE-C\u0026sup2;Q achieves up to 3.6\u0026times; model compression, \u0026lt;\u0026thinsp;1% accuracy degradation, and ~\u0026thinsp;40% energy savings compared to conventional post-training quantization techniques. Furthermore, the compressed models exhibit improved inference latency and memory utilization on ARM-based edge hardware. Unlike traditional pipelines, SyE-C\u0026sup2;Q integrates entropy-guided quantization directly into the compression loop, establishing a new benchmark in the design of resource-efficient, deployable deep learning systems.\u003c/p\u003e","manuscriptTitle":"Synergistic Entropy Coding and Quantization Enable Efficient on Device Neural Networks","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-12-04 18:39:58","doi":"10.21203/rs.3.rs-8018121/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"e5eb3337-1c14-4797-934b-b6fc5fb18c14","owner":[],"postedDate":"December 4th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-23T15:40:03+00:00","versionOfRecord":[],"versionCreatedAt":"2025-12-04 18:39:58","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8018121","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8018121","identity":"rs-8018121","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}